215 results about "Dynamic filtering" patented technology

The invention discloses an unmanned aerial vehicle image small target detection method based on dynamic filtering and an adaptive sparse Transform. According to the method, an end-to-end target detection framework is adopted, a dynamic filtering module is introduced into a backbone network, global feature interaction is achieved through data-dependent frequency domain operation, and linear calculation complexity is maintained. For feature interaction in a scale, an adaptive sparse Transform module is introduced to enhance the capability of focusing key information on high semantic hierarchy features of a model, and noise interference and feature redundancy are effectively suppressed at the same time. Through the combination of dynamic filtering and adaptive sparse Transform, the model can extract image foreground information more effectively on the premise of not significantly increasing the calculation burden, and the problem that a traditional target detection model is susceptible to complex background interference is significantly relieved.

The invention relates to the technical field of automatic production control and industrial data processing, in particular to a PLC-based gypsum-based flame-retardant plate production line control system and method, and the system comprises a data collection module which collects sensor data in real time; the ideal state construction module is used for constructing a pure ideal state vector; the disturbance simulation module is used for generating a theoretical damaged state vector corresponding to a specific fault mode; the difference calculation module is used for generating real and theoretical deviation vectors; the coupling verification control module is used for calculating a similarity value between the actual deviation vector and the theoretical deviation vector; if the similarity value is larger than a preset judgment threshold value, it is judged that a real physical fault exists at present, and a PLC precise compensation instruction is generated and sent to a production line execution mechanism; otherwise, judging that non-physical noise exists at present, keeping the control parameters of the production line unchanged, and starting a dynamic filtering program; according to the method, the problem that physical faults and sensor clutters are difficult to distinguish in a high-noise environment is effectively solved, and missing report is avoided.

The invention relates to the technical field of power distribution networks, and discloses a power distribution network fault self-recovery method and system based on machine learning, and the method comprises the steps: collecting the multi-source fault feature data of a power distribution network in real time, including the voltage and current waveform distortion feature, the load sudden change feature and the equipment temperature anomaly feature; and inputting the multi-source fault feature data into a multi-stage classification model for processing, and by setting a pseudo fault filtering module, establishing a typical interference feature library and setting dynamic filtering thresholds for different working conditions during real-time fault detection of the power distribution network, so that the recognition accuracy of various interference signals is ensured, and the fault detection accuracy of the power distribution network is improved. Meanwhile, similarity comparison is carried out on real-time collected data and pseudo fault features, whether the problems of signal distortion and noise pollution occur at the fault positioning position of the power distribution network or not can be detected in real time, the accuracy of fault type recognition is ensured, the misjudgment risk is reduced, and when strategy adaptation is abnormal, the matching degree of the self-healing strategy and the actual state of the power grid is ensured.

The invention relates to the technical field of artificial intelligence and biomedical sensing, and discloses a cognitive disorder risk identification device and method integrating multi-modal physiological data, and the device comprises an EEG module, an eye movement module, an fNIRS module, an edge calculation module and an acceleration sensor; the EEG module collects EEG data, the eye movement module collects eye movement eye-tracing data, and the fNIRS module collects near infrared spectrum fNIRS data and inputs the data to the edge calculation module; the edge calculation module runs the multi-modal space-time attention fusion model to output a cognitive impairment risk assessment result by using the multi-modal space-time attention fusion model; the acceleration sensor operates based on an operation dynamic filtering algorithm of a motion accelerometer to inhibit signal drift caused by head motion. According to the invention, through the modularized head-mounted multi-mode edge device, the physiological data is collected and edge end processing and cognitive disorder recognition and screening are carried out in combination with the edge calculation module, so that early recognition and screening of neurodegenerative diseases can be conveniently and effectively carried out at low cost.

The invention relates to the technical field of the Internet of Things, and discloses an animal physiological feature data acquisition system and analysis method based on the Internet of Things, and the method comprises the following steps: cleaning and standardizing original sensor data, including filling missing data, eliminating noise interference through a dynamic filtering technology, unifying the dimension of multi-source data, and analyzing the data; the method comprises the following steps of: acquiring a Beidou time service time reference, eliminating redundant features, interpolating and aligning asynchronous sampling data to the Beidou time service time reference, correcting a time sequence phase difference of physiological and behavior data through a dynamic path matching technology, and correcting GPS / Beidou positioning drift in combination with Kalman filtering. Through multi-source sensing fusion and an edge-fog-cloud cooperative computing architecture, the effective collection rate of animal physiological data is achieved, abnormal response is delayed and compressed, the flexible energy supply technology and dynamic communication optimization are combined, the endurance and communication success rate is maintained in a pasture, and the false alarm rate is reduced compared with a traditional method; common diseases such as abnormal body temperature, digestive system diseases and the like are accurately warned and covered.

The invention discloses a convolutional neural network information processing method for heart function dynamic monitoring, and belongs to the technical field of medical detection and monitoring. Comprising the following steps that multi-mode electrocardiosignal data and current physiological state parameters of a patient are obtained, and initial configuration of the intelligent heart function monitoring device is obtained; determining the signal quality grade of each signal channel, and generating a dynamic filtering adjustment strategy of the multi-modal signal acquisition module; configuring a feature extraction strategy of a double-branch convolutional neural network module based on the signal features of the target analysis signal segment and the physiological state parameters; performing multi-scale time sequence feature extraction and frequency domain autonomic nerve regulation feature extraction on the target analysis signal segment by using a double-branch convolutional neural network module according to a feature extraction strategy; executing arrhythmia classification, heart rate variability parameter quantification, heart function evaluation grade and graded early warning tasks, and outputting multi-stage intelligent early warning information from normal monitoring, potential risk and abnormal early warning to an emergency state.

The invention relates to the technical field of building information, and provides a knowledge base construction and retrieval method and system based on a multi-source text in the building field, and the method comprises the following steps: a knowledge base construction stage: constructing a multi-dimensional metadata feature vector for a multivariate text based on a standard classification index table; the method comprises the following steps of: converting and segmenting a document, splicing an end clause with all superior title texts by utilizing a context inheritance algorithm to form a text unit with complete semantics, and performing dynamic filtering based on an analyzed query intention and a metadata vector at a user retrieval stage; then, in the screening set, performing fusion calculation on semantic vector similarity, keyword matching degree and authority offset weight based on effectiveness attribute and implementation time, and performing mixed retrieval and reordering on the text units; and finally selecting a text unit according to a sorting result and inputting the text unit into the large language model to generate answers. According to the method, high-precision and high-compliance intelligent retrieval and question answering of building domain knowledge are realized.

The invention provides an NL2SQL database table dynamic screening method and system based on multi-strategy fusion. The method comprises the steps that a preliminary candidate set is generated, a semantic encoder is used for encoding a user natural language problem and table and column information in a database into vectors, the semantic similarity is calculated, a preset number of tables are screened out from a whole database table based on the similarity, and a preliminary candidate table set is formed; performing deep analysis and reordering, performing multi-dimensional analysis on the tables in the preliminary candidate table set, and calculating a graph relationship compactness score and a content correlation score of each table; and performing fusion decision, inputting the semantic similarity score, the graph relationship compactness score and the content correlation score of each table into a fusion model to obtain a final comprehensive score of each table, and screening out a final correlation table set based on the final comprehensive score. The method has high accuracy and high recall rate, the generation speed and response efficiency of the NL2SQL system are improved, meanwhile, the robustness is high, and the practicability and generalization ability of the system are improved.

The invention relates to the technical field of power transmission line sag measurement, in particular to a power transmission line sag measurement method based on laser point cloud data, which comprises the following steps: controlling a laser radar module and a multispectral imaging module to perform time-space synchronous data acquisition on a power transmission line area, and generating an auxiliary three-dimensional point cloud through a data fusion and compensation module; fusing the laser point cloud and the auxiliary three-dimensional point cloud by adopting an adaptive weighted average algorithm to generate a fused point cloud; performing dynamic filtering processing on continuous multi-frame point cloud data through a dynamic filtering module based on Kalman filtering and an established conductor dynamic model, filtering out high-frequency vibration components, and extracting a steady-state position point cloud of the conductor; the sag value of the power transmission line is calculated through the sag calculation module, so that cooperative sensing and data adaptive fusion of the multi-source sensor in a severe environment are completed, recursive filtering processing is performed on continuous multi-frame point cloud data, and a more real and more accurate data basis is provided for sag calculation.

The invention provides a high-fidelity anti-compression image watermarking method and system based on spectrum-airspace decoupling, and belongs to the field of information security. Firstly, the watermark information is mapped and remodeled; a watermark encoder based on multi-granularity spectrum-spatial domain feature decoupling is constructed, discrete cosine transform is introduced to filter out high-frequency components, Haar wavelet transform is adopted to realize lossless downsampling, a fast Fourier transform dynamic filter is combined to capture global semantic features, and local texture details are combined through multi-scale spatial domain volume accumulation; designing a physical perception and visual self-adaptive dual embedding strategy, and anchoring watermark energy to an anti-compression brightness channel; constructing an anti-attack layer containing differentiable JPEG compression simulation and mixed noise simulation, and participating in network training; and constructing a decoder and designing a loss function to carry out network optimization. According to the method, the robustness of the watermark under strong compression and complex black box attacks is improved, and extremely high visual imperceptibility is realized through physical and visual constraints.

The invention relates to the technical field of audio signal processing, and provides a self-adaptive frequency compensation system for small-sized multichannel sound equipment and a control method thereof, and the self-adaptive frequency compensation system for the small-sized multichannel sound equipment comprises an audio signal acquisition module, a frequency analysis module and a dynamic filtering compensation module which are electrically connected with one another in a communication manner; the audio signal acquisition module is used for acquiring an audio signal at a preset sampling rate through the ADC module; the frequency analysis module is used for executing FFT (Fast Fourier Transform) on the acquired audio signals to calculate energy distribution of corresponding frequency bands and identify missing or attenuated frequency ranges; and the dynamic filtering compensation module is used for selecting an FIR (Finite Impulse Response) or IIR (Infinite Impulse Response) filter according to a frequency analysis result, dynamically adjusting a filter coefficient and a gain value, performing real-time filtering processing on the audio signal and outputting a compensated audio signal. The missing frequency is identified and compensated by performing analysis, filtering, equalization and other processing on the audio signal.

The application discloses a structure point cloud data multi-scale filtering method considering environmental dynamic influence, which firstly adopts a 'K' nearest field method considering structure dynamic influence to perform dynamic filtering processing on original point cloud data, removes scattered outliers in the original point cloud data, and then divides the point cloud data into flat regions and mutation regions based on an improved principal component analysis algorithm (LMSR-PCA), adopts statistical filtering based on local surface fitting for the flat regions, and adopts spatial adaptive bilateral filtering for the mutation regions. The application is used to solve the problems that the existing point cloud data processing method cannot efficiently process massive point cloud data with complex curve characteristics, and cannot solve the problem of excessive smoothing caused by the loss of edge features in the noise reduction process of point cloud data with complex curve characteristics.

The invention relates to the technical field of digital integrated circuit verification, in particular to a large-scale chip post-simulation log file intelligent analysis method and device and a medium. The method comprises the following steps: performing data cleaning on a post-simulation log file; dividing the cleaned log file into a plurality of sub-file blocks according to timestamps, starting multiple threads to process each sub-file block in parallel, counting the types and quantity of time sequence violations, extracting related logs and generating a time sequence violation file; classifying and analyzing the time sequence violation file by utilizing a machine learning algorithm; dynamically filtering the time sequence violation file according to a preset rule, screening key violation information, and repairing time sequence violation caused by signal delay; and in combination with a chip circuit structure and a signal transmission path, positioning a root module or node of a time sequence violation, formulating an optimization scheme and verifying a repair effect. According to the method, efficient and accurate analysis of the post-simulation log file of the large-scale chip is realized, and the efficiency and accuracy of chip verification are improved.

The invention discloses a visual Token compression method and device based on double-path fusion, and the method comprises the steps: obtaining an initial visual Token through employing a visual encoder; obtaining a learnable query vector, and adopting an attention mechanism to enable the learnable query vector to interact with the initial visual Token to obtain a plurality of context Token; analyzing a self-attention map of the initial visual Token in the visual encoder, so as to screen out a plurality of detail Token according to the self-attention map; obtaining a text prompt Token, and calculating a task correlation score of each detail Token by taking the text prompt Token and the plurality of context Token as a query; dynamically filtering or weighting the plurality of details Token according to the task correlation score to obtain a plurality of filtered details Token; the plurality of filtered detail Tokens are spliced with the plurality of context Tokens, so that a compressed visual Token can be obtained; therefore, the key global and local visual information can be intelligently reserved while the high compression ratio is realized and the calculation cost is reduced.

The invention discloses a centrifugal machine dynamic filtering efficiency improving method based on artificial intelligence optimization control, and relates to the technical field of centrifugal machine filtering efficiency optimization. Constructing an edge-cloud collaborative architecture system; the sensing layer deploys various sensors to acquire equipment operation data; the edge calculation layer processes the data and makes a decision; the network transmission layer ensures reliable transmission of data; the cloud platform is used for model training and global optimization; through preprocessing and feature extraction of collected data, the state is evaluated by using a hybrid model fusing deep learning and physical simulation, control parameters are optimized based on a multi-objective optimization function in combination with fuzzy logic and an adaptive algorithm, and dynamic feedback adjustment is realized by using electro-hydraulic servo and machine vision. According to the technology, intelligent operation of the centrifugal machine is achieved, compared with traditional control, the filtering efficiency is improved, energy consumption is reduced, the service life of filter cloth is prolonged, equipment fault diagnosis is more accurate, maintenance response is faster, production benefits are improved, and product market competitiveness is enhanced.

The invention discloses a dynamic filtering and beamforming method for improving the anti-interference capability of a secondary radar. The method comprises the following steps: initializing original data; periodically monitoring the interference congestion degree of the downlink response signal of the secondary radar system in the current radio environment, and judging the interference intensity; dynamically adjusting the transmitting power and the PRF of an uplink interrogation signal of the secondary radar system; notifying the air target responder to reduce the sending rate and the transmitting power of a downlink response signal through the downlink control instruction; the general orientation of the interference source is determined in combination with external monitoring information, adaptive beamforming is performed based on the array antenna, and null is formed in the general direction of the interference source to realize interference suppression of a spatial domain; and performing dynamic filtering processing on the receiving signal after beam forming processing, extracting an effective target response signal sequence, and reconstructing the effective target response signal sequence into complete target response information data according to a predetermined protocol format. According to the invention, the accuracy and reliability of target signal detection and the anti-interference performance of the system are improved.

The invention relates to the technical field of safety monitoring of power transmission equipment, in particular to an intelligent early warning nut and a monitoring method thereof.The nut comprises a quantitative anti-interference sensing and dynamic filtering module, a verification parameter is dynamically calculated based on an interference-verification quantitative model by synchronously collecting loosening distance, vibration and electromagnetic interference data, and the verification parameter is obtained; a real loosening event is accurately identified; the environment coupling trend quantitative prediction module is used for fusing multi-environment factor dynamic weighting such as humidity, wind speed and temperature difference after real loosening is judged, and predicting a future loosening distance through an environment-loosening quantitative correlation model; and the scenarized risk grading response module is used for calculating a scenarized risk quantized value R and triggering grading early warning in combination with the prediction result and the structural importance weight of the bolt mounting position, and generating a strategy instruction containing a maintenance priority. According to the invention, the problems of high false alarm rate, incapability of predictive maintenance and short endurance in the prior art are solved, and accurate sensing, intelligent prediction and graded early warning of the bolt loosening state are realized.

The invention discloses a security risk dynamic filtering and feedback control method for large model output content. The method comprises the step of executing a multi-dimensional risk feature library construction operation. According to the method, semantic features are extracted by adopting a BERT-BiLSTM hybrid model, risk modes are automatically found in combination with an improved DBSCAN algorithm, the updating efficiency of a feature library is improved by 40%, feature complementation is achieved through a text / image / voice three-mode heterogeneous network design and a cross-mode attention mechanism, the composite violation content detection accuracy rate reaches 92.3%, and the method has the advantages that the method is simple and convenient to operate and high in practicability. A static channel is adopted to ensure basic rule adherence, a dynamic channel is adopted to capture context risks, a three-level filtering strategy is adopted, illegal content is clear to be completely shielded within 0.3 second, a visual warning label is added to suspected content, edge content is manually rechecked, and the time is shorter than or equal to 5 minutes, so that when the model is output, content safety is ensured, and the probability of illegal content detection is lowered. And the application value of the large model is furthest maintained.

The invention relates to the technical field of earthwork measurement, in particular to an accurate earthwork measurement system based on a laser radar technology. The system comprises a mobile scanning acquisition module, a point cloud intelligent processing module and an engineering quantity accurate calculation and visualization module. The mobile scanning acquisition module dynamically acquires original three-dimensional point cloud data of a construction area through a laser radar; the point cloud intelligent processing module separates ground point cloud from the complex scene point cloud with high precision through a dynamic filtering algorithm; and the project amount accurate calculation and visualization module accurately calculates the total amount of filling and excavation through a boundary adaptive calculation method. According to the invention, a dynamic filtering algorithm and a boundary adaptive calculation method fusing geometric features and physical attributes are provided, and full-automatic, high-precision and visual measurement of the earthwork volume is realized.

The invention provides a test management method, system and device, and a storage medium. The method comprises the following steps: synchronizing incremental test data of a test platform and defect associated data of a defect management system by setting a period; dynamically filtering invalid cases by adopting a preset cleaning rule; extracting defect numbers of the failed use cases based on a regular expression, and generating a hyperlink defect use case list containing skipping to a defect system; and periodically generating a test report according to the case execution state and the number. Therefore, the test data and the defect associated data of the defect management system are increased through the periodic synchronization test platform, automatic association of cross-system data is realized, the problem that errors are easy to occur in manual matching is avoided, the delay of the test report is reduced through generation of the periodic test report, and second-level defect positioning is realized through hyperlink configuration.

The invention relates to the technical field of natural language processing, in particular to a long text matching method and system based on key information and difference characteristics, and solves the problems that in the prior art, key information is dispersed due to noise interference in long text processing, and semantics of words or phrases in long text processing are more fuzzy and diversified. According to the method, a long text is preprocessed to obtain a test set sentence pair, a training set sentence pair and a verification set sentence pair, then a training model is obtained through sentence-level information entropy screening, word-level dynamic filtering, semantic difference enhancement and adaptive feature fusion, and finally after verification is conducted through the verification set sentence pair, the test set sentence pair predicts and outputs a result. The system comprises a long text matching system data preprocessing unit, a long text matching system function module training unit and a long text matching system function module result output unit. Key information is extracted for learning in long text matching, a large amount of computing power does not need to be consumed, and the matching effect is improved.

The invention discloses an underwater autonomous mapping method based on target detection and deep semantic key point extraction, and the method comprises the steps: firstly, enabling an autonomous system to obtain environment information through trinocular vision and active sonar; secondly, MEK-SLAM is utilized to fuse data and estimate position and attitude, a physical-deep learning framework is combined to construct a double-branch network, and interference of underwater images is reduced; then, the pose of the corrected and fused data is updated in real time through MEK-SLAM, and a stable basis is provided for feature extraction; thirdly, a real-time instance segmentation algorithm is combined with dynamic filtering of time sequence perception, interference of dynamic objects is eliminated, and positioning is carried out only based on static environment characteristics; and finally, adopting a deep learning-based feature detection and description algorithm and a multi-modal feature robust descriptor to optimize feature point matching, and generating a three-dimensional underwater map. According to the invention, the robustness of the autonomous system in a complex dynamic environment is improved; and meanwhile, the image registration precision is improved, error accumulation is reduced, and a high-precision three-dimensional underwater map is generated.

The embodiment of the invention relates to an automatic driving blind area sensing early warning method and system based on radar and vision fusion. The method comprises the steps that point cloud data and image data of a vehicle blind area are acquired; adaptive dynamic filtering is carried out on the point cloud data, point cloud features after filtering are extracted, enhancement processing is carried out on the image data, and visual features after enhancement are extracted; performing spatial alignment on the visual features and the point cloud features, and performing weighted fusion on the aligned visual features and the point cloud features based on a gating weight to obtain a fusion perception result; and judging whether a potential dangerous target exists in the blind area of the vehicle or not according to the fusion sensing result, and executing graded early warning according to the category information and the motion state information of the potential dangerous target. According to the embodiment of the invention, advantages of radar and vision are complemented, the method is suitable for a severe environment, and deep multi-modal fusion is carried out on the point cloud features and the visual features for perception and early warning, so that the safety of automatic driving is improved.

A computer-implemented method performs dynamic filtering within a service mesh. While transferring data from one microservice to another microservice, each microservice evaluates a level of trust with each upstream and downstream microservice based on the level of trust score. Each microservice evaluates which data should be filtered while transferring data to downstream microservice and which data should be filtered while receiving data from upstream microservice.

The invention discloses a UWB radar multi-target sensing and tracking method and system. The method comprises the steps that multi-scene original echo signals are collected through a UWB radar, a synchronous optical motion capture system obtains real position information of a target, data diversity is enhanced through noise injection and time-frequency transformation, and a multi-modal data set is constructed; de-noising and signal enhancement are carried out by adopting a three-stage cascade processing flow, and features are extracted by utilizing a CNN and Transform hybrid encoder; based on a graph neural network and a dynamic filtering theory, constructing a model of an improved PointPill detection head and a GAT tracking head to carry out target detection and track association; and finally deploying to edge equipment to realize real-time multi-target sensing and tracking by combining a multi-task joint loss function with a strategy optimization model such as adaptive loss balance. According to the method, the traditional multi-sensor dependence is broken through, the track continuity in a complex scene is improved in a target shielding scene, the probability of wrong tracking and missing tracking is reduced, and high-precision and high-real-time multi-target sensing and tracking are realized in the complex scene.

The invention relates to the technical field of remote sensing image processing, in particular to a raft laver culture area end-to-end extraction method based on frequency dynamic filtering, and the method comprises the steps: obtaining an original remote sensing image of a raft laver culture area, carrying out the preprocessing, constructing a mask data set for high-level semantic supervision, and carrying out the extraction of the original remote sensing image; constructing a hierarchical supervision neural network model, and performing model training on the constructed hierarchical supervision neural network model, including constructing a composite loss function composed of mask loss, gravitational field loss, vertex classification loss and vertex offset loss; and performing optimization processing on a prediction result output by the hierarchical supervision neural network model, and generating a breeding area contour vector result based on the optimized binary mask and vertex coordinates. The method solves the problem that the extraction precision is suddenly reduced in a scene of relatively strong spatial spectrum characteristic heterogeneity of the breeding area in a traditional method, and improves the extraction accuracy of the breeding area. And reliable technical support is provided for large-area and high-precision monitoring of an offshore raft type laver culture area.

The invention discloses an automatic counting information synchronization method suitable for multiple types of cargo trays, and relates to the technical field of intelligent warehouse management, and the method comprises the following steps: inputting a single-piece standard weight parameter of a target cargo in an industrial touch screen, and binding the cargo information with a tray identifier through bar code recognition; goods to be measured are sequentially placed on the weighing platform provided with the four-point symmetrically-distributed weighing modules, the weighing modules collect total weight data of the goods in real time, and impact signals generated in the weighing process are dynamically filtered. According to the invention, the weighing precision is improved through structure optimization and dynamic filtering, automatic and accurate counting of the number of goods is realized in combination with a residual error correction algorithm, the security and completeness of data synchronization are guaranteed by adopting an encrypted two-dimensional code technology, real-time updating and traceable management of inventory information are realized, and the system is convenient to use. And the stability and the management efficiency of the intelligent warehousing system are comprehensively improved.

The invention discloses a multi-mode adaptive weight SLAM method and system for a construction site. The method comprises the steps that a mapping model is established through sensor joint calibration, laser radar, camera and IMU data are synchronously collected through hardware, and dynamic filtering, distortion compensation and feature extraction are carried out on point clouds and images; the voxel entropy, the angular resolution and the feature density are calculated in real time to evaluate the degradation degree of each mode, and a visual residual error is dynamically adjusted by adopting an adaptive weight and frame skipping mechanism to ensure reasonable information injection in a degradation scene; iMU pre-integration, LiDAR point-surface residual errors and weighted visual reprojection residual errors are sequentially fused in an incremental Kalman sliding window frame, an automatic repositioning strategy is matched, and a high-precision pose and dense color point cloud map is output. According to the method, calibration, synchronization, degradation detection, sliding window optimization, map release and relocation are integrated, and the mapping precision and robustness can be remarkably improved under the complex working conditions of dust shielding, illumination variation, dynamic obstacles and the like.

The invention discloses an underwater salient target detection method and system based on a double-flow fusion network, and belongs to the technical field of computer vision. The method comprises the following steps: respectively extracting multi-scale features of an RGB image and a depth image through a double-flow encoder; in the shallow layer, fusing and enhancing the edge and detail information of the bimodal features through an edge fusion module; in a deep layer, content-adaptive cross-modal semantic fusion is realized in a frequency domain through a dynamic filtering module; fusing the multi-scale features through a cross-layer aggregation decoder to generate a rough saliency map; extracting detail features from the original RGB image through a global detail purification network; and finally, fusing the rough saliency map and the detail features, and outputting an underwater saliency target prediction map. The objective of the invention is to improve the precision and boundary definition of salient target detection in an underwater complex scene.

The invention discloses an infrared and visible light image fusion method based on degradation perception and frequency integration, belongs to the technical field of image fusion, is used for image fusion, and comprises the steps of obtaining an infrared image and a visible light image, inputting the infrared image and the visible light image into an encoder, carrying out fast Fourier transform, obtaining a phase and an amplitude, and respectively carrying out dynamic filtering. Performing channel splicing on the dynamic filtering results of the two phases by using a nonlinear mapping module to obtain a fused phase spectrum, and performing inverse fast Fourier transform on the fused phase spectrum and the dynamic filtering result of the amplitude to obtain a fused feature; inputting the fusion features into a degradation perception expert prompt module, carrying out feature multiplication on the fusion features and scaling factors, then carrying out feature addition on the fusion features and bias parameters, completing modulation of the fusion features, and inputting the fusion features into an encoder to obtain a fusion image; according to the method, the infrared and visible light image fusion quality is remarkably improved, the expressive power of a structural sensitive frequency band is enhanced, key texture and geometric information are effectively reserved, and the structural fidelity of the fused image is remarkably improved.